3b. Feature Matching¶

The feature matching takes the reference data and attempts to identify corresponding features in the OSM data set. Feature matching is a necessary precondition to compare single features rather than feature characteristics on study area a grid cell level.

Method

Matching features in two road data sets with each their way of digitizing features and a potential one-to-many relationship between edges (for example in the case where one data set only maps road center lines, while the other map the geometries of each bike lane) is not a trivial task.

The method used here converts all network edges to smaller segments of a uniform length before looking for a potential match between the reference and the OSM data. The matching is done on the basis of the buffered distance between objects, the angle, and the undirected Hausdorff distance, and is based on the works of Koukoletsos et al. (2012) and Will (2014).

Based on the matching results, the following values are computed:

  • The number and length of matched and unmatched edges, in total and per grid cell
  • A comparison of the attributes of the matched edges: Is their classification of cycling infrastructure as protected or unprotected the same?

Interpretation

It is important to visually explore the feature matching results, since the success rate of the matching influences how the analysis of number of matches should be interpreted.

If the features in the two data sets have been digitized differently - e.g. if one data set has digitized bike tracks as mostly straight lines, while the other includes more winding tracks, the matching will fail. This is also the case if they are placed too far from each other. If it can be confirmed visually that the same features do exist in both data sets, a lack of matches indicates that the geometries in the two data sets are too different. If however it can be confirmed that most real corresponding features have been identified, a lack of matches in an area indicates errors of commission or omission.

Sections

  • Match features
    • Run and plot feature matching
    • Matched and unmatched features
    • Feature matching summary
  • Analyze feature matching results
    • Matched features by infrastructure type
    • Feature matching success
  • Summary

Match features¶

Run and plot feature matching¶

Segments created successfully!


Starting matching process using a buffer distance of 15 meters, a Hausdorff distance of 17 meters, and a max angle of 30 degrees.


Buffer matches found! Continuing with final matching process...


60809 reference segments were matched to OSM edges
2886 reference segments were not matched
Feature matching completed!
Make this Notebook Trusted to load map: File -> Trust Notebook
Interactive map saved at results/COMPARE/cph_geodk/maps_interactive/segment_matches_15_17_30_compare.html

Matched and unmatched features¶

Feature matching summary¶

Edge count: 2585 of 5669 OSM edges (45.6%) were matched with a reference edge.
Edge count: 4313 out of 4705 GeoDanmark edges (91.67%) were matched with an OSM edge.
Length: 422.15 km out of 811.53 km of OSM edges (52.02%) were matched with a reference edge.
Length: 601.10 km out of 626.48 km of GeoDanmark edges (95.95%) were matched with an OSM edge.
2023-01-18T14:54:15.202642 image/svg+xml Matplotlib v3.6.2, https://matplotlib.org/
2023-01-18T14:54:15.296560 image/svg+xml Matplotlib v3.6.2, https://matplotlib.org/

Analyze feature matching results¶

Matched features by infrastructure type¶

Feature matching success¶

In the plots below, the count, percent, and length of matched and unmatched features in each data set are summarized.

Warning

The number of matched features in one data set in a grid cell does not necessarily reflect the number of matched features in the other data set, since an edge can be matched to a corresponding edge in another cell. Moreover, the local count refers to edges intersected with the grid cell. For example, a long bike lane crossing 3 cells will thus be counted as matched in 3 different cells. This does not change the relative distribution of matched/unmatched features, but it does entail that the overall summary of matched/unmatched features above uses a different total count of edges than the plots below.

Summary¶

Feature Matching Results
  OSM GeoDanmark
Count of matched edges 2,585 4,313
Percent matched edges 46% 92%
Length of matched edges (km) 422 601
Percent of matched network length 52 96
Local min of % matched edges 3% 7%
Local max of % matched edges 100% 100%
Local average of % matched edges 72% 96%